PAIRED WATERSHED ANALYSIS WITH A SIMULATED CONTROL SITE A Thesis

This thesis provides a rigorous statistical analysis of whether agricultural best management practices (BMPs) have reduced total dissolved phosphorus (TDP) loads from dairy farms in the Cannonsville Reservoir basin in New York State, a drinking water supply for New York City. This paper introduces the use of a simulation model as a control site in a paired watershed analysis as a technique that can be applied in basins when a control site is not available or feasible. In our analysis, the simulated control site also plays a critical role in allowing the analysis to separate the effects of BMPs from those of other water quality interventions. A variety of univariate and multivariate statistical models are considered along with different model structures. In addition to paired watershed models, several single watershed multivariate regression models are considered. Statistical power and the minimum detectable treatment effect (MDTE) are computed. Multivariate paired watershed models were clearly superior to the univariate paired watershed models recommended by EPA in terms of adjusted R and ability to detect statistically significant BMP treatment effects. Modifications to the traditional log-linear regression model structure allowed the analysis to control for wastewater treatment plant (WWTP) upgrades and changes in the dairy cow population, using a physically meaningful statistical model. The results document a 41.5% reduction in TDP loads from non-point sources associated with the implementation of BMPs in the Cannonsville basin. Overall, this study shows that paired watershed analysis with a simulated control site is an innovative statistical method which should have great value in many applications.